A bird-like skull in a Triassic diapsid reptile increases heterogeneity of the morphological and phylogenetic radiation of Diapsida

The Triassic Period saw the first appearance of numerous amniote lineages (e.g. Lepidosauria, Archosauria, Mammalia) that defined Mesozoic ecosystems following the end Permian Mass Extinction, as well as the first major morphological diversification of crown-group reptiles. Unfortunately, much of our understanding of this event comes from the record of large-bodied reptiles (total body length > 1 m). Here we present a new species of drepanosaurid (small-bodied, chameleon-like diapsids) from the Upper Triassic Chinle Formation of New Mexico. Using reconstructions of micro-computed tomography data, we reveal the three-dimensional skull osteology of this clade for the first time. The skull presents many archaic anatomical traits unknown in Triassic crown-group reptiles (e.g. absence of bony support for the external ear), whereas other traits (e.g. toothless rostrum, anteriorly directed orbits, inflated endocranium) resemble derived avian theropods. A phylogenetic analysis of Permo-Triassic diapsids supports the hypothesis that drepanosaurs are an archaic lineage that originated in the Permian, far removed from crown-group Reptilia. The phylogenetic position of drepanosaurids indicates the presence of archaic Permian clades among Triassic small reptile assemblages and that morphological convergence produced a remarkably bird-like skull nearly 100 Myr before one is known to have emerged in Theropoda.

The Triassic Period saw the first appearance of numerous amniote lineages (e.g. Lepidosauria, Archosauria, Mammalia) that defined Mesozoic ecosystems following the end Permian Mass Extinction, as well as the first major morphological diversification of crown-group reptiles. Unfortunately, much of our understanding of this event comes from the record of large-bodied reptiles (total body length > 1 m). Here we present a new species of drepanosaurid (small-bodied, chameleon-like diapsids) from the Upper Triassic Chinle Formation of New Mexico. Using reconstructions of micro-computed tomography data, we reveal the three-dimensional skull osteology of this clade for the first time. The skull presents many archaic anatomical traits unknown in Triassic crown-group reptiles (e.g. absence of bony support for the external ear), whereas other traits (e.g. toothless rostrum, anteriorly directed orbits, inflated endocranium) resemble derived avian theropods. A phylogenetic analysis of Permo-Triassic diapsids supports the hypothesis that drepanosaurs are an archaic lineage that originated in the Permian, far removed from crowngroup Reptilia. The phylogenetic position of drepanosaurids indicates the presence of archaic Permian clades among Triassic small reptile assemblages and that morphological convergence produced a remarkably bird-like skull nearly 100 Myr before one is known to have emerged in Theropoda.

Diagnosis
Specimens for anatomical comparisons are listed in appendix C. A drepanosaurid diapsid differing from Hypuronector limnaios, Megalancosaurus preonensis and Vallesaurus cenensis (the only drepanosauromorphs with skull material) in the complete absence of teeth, a dorsoventrally taller retroarticular process with a triangular shape in lateral view, and cervical neural spines with subequal anteroposterior lengths and transverse widths.

Comparative anatomy
The identification of this specimen as a drepanosaurid is based on its cervical vertebral anatomy. Drepanosaurids possess heterocoelous cervical vertebral centra with saddle-shaped articular surfaces. The prezygapophyseal facets face anteriorly and extend far anteriorly relative to the anterior margin of the centrum. The neural spines are anteroposteriorly short and strongly inclined anterodorsally. In each of these features, Av. renestoi is very similar to drepanosaurids with cervical series, specifically Drepanosaurus unguicaudatus [27,31]. The bones of the skull are loosely articulated with one another, much as in the few other known drepanosauromorph skulls. It is similar in size to the known skulls of Me. preonensis (approx. 27 mm) and substantially larger than the skull of the holotype of V. cenensis (approx. 16 mm).

Bird-like traits
The skull of Av. renestoi exhibits a number of striking similarities to avian theropods (figures 2 and 3). The rostrum is slender and acuminate, as has been noted in the Italian drepanosaurid Me. preonensis [39,40]. Avicranium renestoi combines this shape with a completely edentulous rostrum and palate (figure 3e). The construction of the orbit differs from that in most Triassic diapsids, in which the cavity is directed anterolaterally [33,41,42]. In Av. renestoi, the frontal, postfrontal and postorbital all contribute to a transversely broad postorbital septum, which directs the orbital cavity anteriorly (figure 3c). The analogous postorbital process in maniraptorans integrates processes of the frontal, squamosal and laterosphenoid [43]. In most birds, the process is formed primarily by a cartilaginous expansion of the laterosphenoid. Renesto & Dalla Vecchia [40] also suggested binocular vision for Me. preonensis, based on the tapering rostrum and broadened orbital and temporal regions. The endocranium preserves some of the most striking departures of the Av. renestoi from other Triassic reptiles. The contribution to the braincase of the paired frontal and parietal bones is both broad transversely and tall dorsoventrally. This contribution is so prominent that the contributing portion of the frontal is domed dorsally well above the orbital margin (figure 3a). This corroborates the hypothesis by [40] that the Italian drepanosaurid Me. preonensis had an inflated, 'bulging' skull roof [39, p. 251]. Among diapsid reptiles, a similar shape otherwise only occurs in maniraptorans [44][45][46][47] and some pterosaurs [48,49], taxa that possess enlarged brains relative to other Mesozoic diapsid groups. The reconstructed dorsal surface of the endocast of the Av. renestoi resembles those of Pterosauria and maniraptorans in that the cerebrum is large and broad, occupying much of the anteroventral length of the frontal [49][50][51]. An additional large lobe is formed by the posterior portion of the parietal, likely the optic lobes (based on comparisons with Alligator mississippiensis and Gallus domesticus in [52]). The anterior outlet of the osseous braincase in Av. renestoi is also transversely broad; the prootics angle strongly medially at their anterior tips to meet dorsolaterally inclined clinoid processes of the parabasisphenoid. A brain enlarged in the way this endocast suggests is otherwise unknown in a Triassic reptile (figure 3a). Past studies correlate the enlargement of the brain in pterosaurs and maniraptorans with the adaptation of those taxa to flight-it may be that the enlargement of the drepanosaurid brain followed a similar path to an adaptation to the three-dimensional environments required by arboreality, precision grasping and enhanced stereoscopic vision [29,40,49,50,53]. The inclination of the occipital condyle relative to the long axis of the skull is unclear, owing to the disarticulation of the Av. renestoi holotype. In our reconstruction, the occipital condyle is slightly posteroventrally inclined relative to the long axis of the skull, in contrast to the strong posteroventral inclination described for Me. preonensis [40]. However, the distortion of the skull of Av. renestoi may obscure the original shape of the craniocervical articulation.

Plesiomorphic traits
The anatomy of the suspensorium and other morphologies of the braincase stand in stark contrast to the 'advanced' features of the skull roof and rostrum. The squamosal is a dorsoventrally tall, anteroposteriorly broad bone. It exhibits both lateral and posterior laminae that frame the quadrate on those sides, as in archaic eureptiles (e.g. Captorhinus aguti [43,54]) and diapsids (e.g. Araeoscelis gracilis [55]). This contrasts with the condition in younginiform and saurian reptiles (appendix C), in which the quadrate is only framed laterally. In younginiform and saurian taxa, the quadrate also extends dorsally to fit into a fossa on the ventral surface of the squamosal-a feature absent in Av. renestoi. The quadrate itself is dorsoventrally short and vertically oriented, lacking the posterior embayment in most early saurian reptiles (figure 3f ) [33,[56][57][58].
The braincase exhibits a number of traits more commonly found in non-saurian diapsids. The occipital condyle exhibits a deep, posterior depression (=notochordal pit) across much of its surface and the basal tubera barely extend ventrally below the condyle (similar to Ca. aguti [59], Ar. gracilis [55]) (figure 3d). The foramen ovale is extremely large and extends to the ventralmost margin of the braincase. The stapes is massive, with a footplate that entirely fills the foramen and a lateral stem that is larger in all dimensions than the paroccipital process of the opisthotic (figure 3d). Foramina ovale and stapedes of this great size are common in early amniotes [59,60], but they are substantially smaller in younginiform diapsids (e.g. Youngina capensis [61]) and early saurians (e.g. Mesosuchus browni [60], Prolacerta broomi [58]). There is no evidence of a laterosphenoid ossification, as in Archosauriformes [62][63][64].
The plesiomorphic diapsid characters of the skull in Av. renestoi strongly suggest a plesiomorphic ear. Extant reptiles possess a tympanic membrane framed anteriorly by the concavity of the quadrate, which medially contacts a cartilaginous extracollumella, which in turn meets a very slender, osseous stapes [65,66]. The absence of an embayed quadrate and tympanic crest in Av. renestoi suggests the absence of a tympanic membrane. The large foramen ovale with prominent contributions by parabasisphenoid and basioccipital is more common in non-younginiform and non-saurian amniotes, as is the large stapes [59,65,66]. Thus, Av. renestoi lacks the major osteological correlates of impedance-matched hearing. An atympanic condition occurs in a number of extant lepidosaurs (e.g. chameleons, Sphenodon), although these taxa exhibit a slender stapes and a condyle-cotyle articulation between quadrate and squamosal and are widely considered to have undergone secondary loss of external ears. The archaic ear morphology in Av. renestoi, in concert with the other plesiomorphic amniote traits discussed above, contrasts sharply with the comparatively 'advanced' condition in most Triassic Sauria [58,60,67].

Phylogenetic analysis
In the light of the extensive new data on the cranial anatomy of Drepanosauromorpha provided by AMNH FARB 30834, we integrated the taxon into a phylogenetic analysis focused on terrestrial Permo-Triassic Diapsida and early Sauria (modified from [10,11,68]  and detailed results in appendix C. In the most-parsimonious trees, Drepanosauromorpha is recovered as an extremely early-diverging clade of Diapsida, occurring outside of a clade including Permian 'younginiform' diapsids and Sauria (figure 4). The oldest-known younginiform diapsid (herein referred to as Tropidostoma Zone Youngina) dates to the lowermost Upper Permian [69], suggesting that the lineage including drepanosauromorphs must have originated by the end of the Middle Permian (approx. 260 Myr). This phylogeny also recovers Kuehneosauridae, typically found as the sister taxon of Lepidosauria in cladistic analyses of early Diapsida (e.g. [70][71][72]), as deeply nested within Archosauromorpha (postulated in [73]).

Discussion and conclusion
These results indicate that drepanosauromorphs represent a deep divergence within Diapsida, earlier than that of crown-group reptiles, but one that persisted through the PTE and radiated deep within the Triassic [11,27]. A number of the non-saurian diapsids included in this analysis are taxa that also survived the PTE (Weigeltisauridae per [74], Hovasaurus boulei per [75]), indicating that the survival of drepanosauromorphs among non-crown-group reptiles was not a unique event.
Our revised phylogeny, combined with the extensive character data provided by the Av. renestoi holotype, strongly supports the hypothesis that Drepanosauromorpha are non-saurian diapsids. Phylogenetic analyses have long recognized that a number of crown-group reptile lineages (mostly early archosauromorphs) had diverged by the PTE, despite their initial appearance in the fossil  [9,10,33,63]. We tested hypothetical placements of drepanosauromorphs among crown-group reptiles through constraint analyses, but found these to be substantially less parsimonious (appendix C). That result, along with the recognition of numerous other non-crown-group lineages within the Triassic indicates that the Triassic diapsid radiation was far more phylogenetically heterogeneous than traditionally realized.
The general bird-like shape of the drepanosaurid rostrum has long been recognized, owing to complete but crushed specimens from the Upper Triassic of Italy. The three-dimensional preservation of AMNH FARB 30834 adds substantially to the bird-like features of the skull, including the frontated orbits and presumed binocular vision, the absence of teeth, possible fusion of the premaxillae and the inflated endocranium. However, these features occur in conjunction with a strikingly plesiomorphic braincase, suspensorium and postcranial skeleton [27]-features that strongly support the hypothesis that these bird-like features are entirely convergent. Bird-like features have been noted in a number of small Triassic diapsids-including Longisquama insignis and the putative stem-bird Protoavis texensis-which have been used to support the hypothesis that key features of the bird skull evolved very early in the Mesozoic [76,77]. This conception of bird evolution stands at odds with the fossil record of Theropoda, which suggests the gradual acquisition of avian cranial features throughout the Jurassic and Cretaceous [78][79][80]. The mosaic anatomy of Av. renestoi instead supports the hypothesis that several bird-like traits first emerged in a Triassic diapsid lineage entirely outside of crown-group reptiles [36]. The brain of Av. renestoi differs greatly from that in most Permian and Triassic diapsids. The cerebrum is substantially wider than the olfactory tracts and the endocranium occupies a substantial proportion of the transverse width of the skull, distinctly similar to the brains of maniraptorans (e.g. [50,81,82]), living birds (e.g. [83][84][85]) and pterosaurs (e.g. [49,86]). Many authors have suggested that the proportional expansion of brain and cerebrum size in these taxa is an adaptation to the sensory complexity required for navigating three-dimensional habitats [87][88][89]. The anteriorly directed orbits in Av. renestoi, coupled with the hypothesized arboreal habitat for drepanosauromorphs [27,39] suggest a complex sensory life for the animal and may explain the similarities in brain shape to flying and arboreal taxa. Further testing of this hypothesis requires better preserved endocasts and reconstruction of the vestibular apparatus of other drepanosauromorphs.
This phylogenetic study, in concert with the bird-like characters of the skull of Av. renestoi, increases the known disparity achieved by terrestrial diapsid reptiles during the Triassic Period and extends the pattern of morphological convergence on later Mesozoic lineages during the Triassic beyond Archosauromorpha into a non-crown-group reptile clade. This and similar discoveries demand constant re-evaluation of the phylogenetic diversity and morphological disparity of fossil groups involved in the recovery from the PTE.
Data accessibility. The CT datasets that formed the basis of this study are archived on Dryad (http://dx.doi.org/10.5061/ dryad.f5q10). The phylogenetic data matrix is available on the Dryad site in TNT format and also on Morphobank (www.morphobank.org) as project 2214. The full list of taxa, characters and phylogenetic methodologies are presented in appendix C. the specimen to be an early lepidosaur, an identification refined to the 'suborder Eolacertailia' in later papers [25, p. 1127]. This referral was made based on general morphological comparisons and precladistic diagnoses of Lepidosauria. Berman & Reisz [30] referred the New Mexican reptile Dolabrosaurus aquatilis from the Petrified Forest Member of the Chinle Formation to Drepanosauridae, which they also considered to be lepidosaurian. In all the aforementioned cases, the authors perceived a mixture of plesiomorphic and derived diapsid characters in the specimens.
At roughly the same time as the description of the Drepanosaurus holotype, Calzavarra et al. [91] described a partial skull and skeleton of a small reptile from the Upper Triassic Dolomia di Forni of Italy as the holotype specimen of Me. preonensis. They referred to the specimen as a 'thecodont' (i.e. early archosauriform). An early phylogenetic study by Evans [73] suggested that Megalancosaurus was closer to 'prolacertiforms', an arrangement of long-necked early archosauromorphs including Protorosaurus and Tanystropheus. It would not be until the mid-1990s, when some specimens referred to Drepanosaurus were correctly recognized as postcranial skeletons of Megalancosaurus, that a close relationship between the putative lepidosaur and 'thecodont' was recognized [26]. Renesto [26] also hypothesized that Megalancosaurus was closely related to 'prolacertiforms'.
In 1993, Feduccia and Wild published the hypothesis that Megalancosaurus was not only an earlydiverging archosaur but also a close relative of birds [77]. They noted a number of cranial features (e.g. a bird-like beak with small teeth; large, bird-like orbit; enlarged braincase) and postcranial features (e.g. slender, strap-like scapula; putative fused clavicles) supporting their hypothesis. This hypothesis was not presented in a rigorous context with a phylogenetic analysis, and the only subsequent study to include both birds and drepanosaurs [36] did not resolve a close relationship between the two.
Systematists have performed numerous cladistics analyses of early diapsid and archosauromorph relationships incorporating a sample of drepanosaurs. In his unpublished PhD thesis analysis, Merck [92] included both Drepanosaurus and Megalancosaurus into a large-scale study of Permo-Triassic Diapsida. He recovered the two in a clade of Archosauromorpha that was sister to a marine reptile clade including Thalattosauria, Sauropterygia and Ichthyopterygia. By contrast, Merck [35] later presented the hypothesis that drepanosauromorphs and weigeltisaurids formed a clade outside of crown-group reptiles. Dilkes [33] corroborated the hypotheses of Evans [73] (for Megalancosaurus) and Renesto [26], by recovering a Drepanosaurus + Megalancosaurus clade as sister to Tanystropheidae within Protorosauria. Numerous authors have modified the Dilkes analysis (e.g. [34,41,93], consistently recovering drepanosauromorphs as nested within Archosauromorpha). Renesto et al. [27] incorporated the other named drepanosauromorphs into a modified Dilkes [33] analysis, which likewise supported the early archosauromorph position.
By contrast, other analyses have recovered drepanosauromorphs in various positions outside of Sauria altogether, placing them as the sister taxon to either the gliding Weigeltisauridae [35,36] or Kuehneosauridae [94]. In his analysis, Senter [36] recovered drepanosauromorphs outside of Sauria in a clade including Weigeltisauridae and the poorly known Middle Triassic diapsid Lo. insignis. He dubbed this clade Avicephalia. Renesto & Binelli [28] critiqued the Senter [36] analysis and reanalysed the matrix. Renesto & Binelli [28] incorporated the pterosaur Eudimorphodon into their analysis, which nested as the sister taxon to drepanosauromorphs within Avicephalia. However, after correcting some errors in the original Senter [36] matrix, the Eudimorphodon + Simiosauria clade was recovered as the sister taxon to Archosauriformes. Renesto et al. [27] made brief reference to the possibility of a close relationship between pterosaurs and drepanosauromorphs.
In a later iteration of the Müller analysis, Bickelmann et al. [95] noted that drepanosaurs acted as a wildcard taxon following the addition of new operational taxonomic units. This phylogenetic instability was attributed to the meager amount of character data coded for drepanosaurs in most analyses, owing to the crushing distortion in nearly all drepanosaur skeletons (e.g. Renesto et al. [27]). A summation of the cladistics hypotheses for the affinities of drepanosauromorphs is presented in appendix A and figure 5.
The absence of a coherent hypothesis for the relationships of this group has implications for interpreting the extreme ecomorphology of drepanosauromorphs and the Permo-Triassic radiation of diapsid reptiles. The hypotheses by Dilkes [33] and Renesto [26] suggest that drepanosaurs are deeply nested among crown-group reptiles within the early archosauromorph radiation, specifically within a clade of long-necked, small-headed 'protorosaurs'. By contrast, the hypotheses of Merck, Senter and Müller suggest that drepanosaurs are not crown-group reptiles, but instead are closely related to archaic Palaeozoic lineages. The members of these lineages are typically smaller in body size, and both hypotheses suggest that the sister taxon of drepanosaurs were extreme gliding specialists. Resolving the ancestry of drepanosauromorphs provides important context for the diversification of small diapsids in the Permian and Triassic. Senter [36] Müller [94]    Shared Material Instrumentation Facility (Durham, NC, USA) by technician James Thostenson. The specimen was scanned at a resolution of 0.0448 mm for a total of 1998 slices (190 kV, 78 mA). The contrast between the bone and matrix is somewhat problematic. Many of the bones contain dense, radiopaque material that appears bright white on the scan slices. These are surrounded by diffuse grey halos. The CT data (in DICOM format) are available for download on Data Dryad under the title 'Data from: A bird-like skull in a Triassic diapsid reptile increases heterogeneity of the morphological and phylogenetic radiation of Diapsida'. The specimen was digitally segmented in VG Studio Max 2.2. Based on the visible bone of the frontal and parietals on the dorsal surface of the black, the grey halos are reflective of the true extent of the bone. As such, clusters of the white material were first segmented. These were then expanded to incorporate the diffuse halos. Figure 1 was rendered in VG Studio Max, using the 'Volume Render (scatter)' option. We then extracted individual bones as STL surfaces.
The model of the rearticulated drepanosaurid skull was constructed using Maya (v. 2016, Autodesk). Disarticulated bones were fitted together using contact surfaces visible in the extracted surface files. Multiple angles on the rearticulated skulls are presented in figure 7. The endocast in figure 3a was reconstructed as a three-dimensional surface using the ventral surfaces of the right frontal and parietal in AMNH FARB 30834. The surface was mirrored and imaged in Maya 2016.

Appendix C. Phylogenetic analysis: data matrix and results
The phylogenetic dataset used here is a combination of Pritchard et al. [68] (and its expansion in Nesbitt et al. [10]) and Pritchard et al. [11] otherwise noted. Bolded characters have been modified in a substantial way from their original use in Pritchard et al. [68] and/or Nesbitt et al. [10]. Bolded and italicized characters are replacements for characters removed from the dataset in Nesbitt et al. [10]. Notes may be found below such characters regarding the grounds for removal. For new or modified characters, references to past or similar usages in other datasets are referenced. 1) Premaxilla external sculpturing: (0) surface is smoothly sculptured, (1) premaxilla is marked by anteroventral striations. 2) Premaxilla, ventral margin, orientation relative to long axis of skull: (0) margin horizontal, roughly inline with maxillary ventral margin; (1) slight downturn, such that the margin trends anteroventrally; (2) extensive downturn, premaxilla extends to ventral margin of dentary. ORDERED.

10) Maxilla, lateral surface near anteroposterior midpoint: (0) marked by subequal neurovascular foramina, (1) bears single neurovascular foramen that is anteroposteriorly longer than all others.
-This character replaces Character 10 of Nesbitt et al. [10], which described the presence of a midline contact of the prefrontal bones. This state is known only in Choristodera, which are not represented in this dataset. It has thus been removed. -This character replaces Character 27 of Pritchard et al. [68] and Nesbitt et al. [10], which described the relative position of the medial process of the postorbital to the postfrontal. We have incorporated this character to distinguish taxa that truly lack a contact between the postorbital and the midline skull roof elements. Character 254 also partly addresses this morphology, describing the shape and anteroposterior dimensions of the dorsal exposure of the postfrontal. -Pritchard et al. [68] and Nesbitt et al. [10] described the presence and absence of the lateral/descending process of the squamosal in their Character 33 and the relative anteroposterior breadth of that process in Character 34. Here, we present a composite of those characters, with a slender lamina seen as an intermediate between the absence of the structure and the anteroposteriorly broad laminae of early diapsids. bars in Proterosuchus specimens lacking quadratojugal anterior processes. As the characters do appear independent, we integrate this character into the present analysis.

40) Quadratojugal, anterior process, shape: (0) paralleling dorsal and ventral borders, (1) anteriorly tapering anterior process.
-This character may be found as Character 39 in Pritchard et al. [68] and Nesbitt et al. [10]. It here replaces the original Character 40 in those datasets, which described the relative dorsoventral height of the quadratojugal. However, first-hand study of certain taxa with supposedly dorsoventrally low quadratojugals (e.g. Ar. gracilis, MCZ 4036) suggests that the full dorsoventral height of the bone is difficult to assess due to the anteroposterior breadth of the squamosal. As such, we have removed that character pending further study of this state in early Diapsida. -We have removed the original Character 40 of Pritchard et al. [68] and Nesbitt et al. [10], which described the position of the quadrate foramen either between quadratojugal and jugal or within the quadrate. In the taxa studied in this analysis, the only species with a purely quadrateenclosed foramen are also those that lack a discrete quadratojugal ossification. As such, we have removed the foramen character to avoid redundancy. -This character describes the development of a deep posterior concavity on the tympanic crest in many Lepidosauria (e.g. G. bridensis [117], U. acanthinura (YPM R 13525)). Taxa coded as '0' for Character 42 are coded as '-' for this character. -Taxa coded as '1' for Character 44 are coded as '-' for this character.

48) Palatine. Lateral tooth row, dental morphology: (0) similar to other palatal teeth; (1) enlarged relative to all other palatal teeth, akin to marginal teeth in size and morphology.
-This character was not present in the datasets of Pritchard et al. [68] nor Nesbitt et al. [10]. The original Character 48 described the presence of teeth on the anterior process of the pterygoid. It is accounted for in Characters 49 and 50 of this study. -This character describes the enlarged palatine teeth of Rhynchocephalia.
-This character is included in Pritchard et al. [68] and Nesbitt et al. [10] as Character 50.
-This character is included in Pritchard et al. [68] and Nesbitt et al. [10] as Character 51.
-This character is included in Pritchard et al. [68] and Nesbitt et al. [10] as Character 52.
-This character is included in Pritchard et al. [68] and Nesbitt et al. [10] as Character 53.
-We have removed Character 54 of the original datasets of Pritchard et al. [68] and Nesbitt et al. [10], which described the shape of the midline space framed by the contacting pterygoids (anteriorly tapered or anteriorly curved). However, the curved shape appears to be the simple by-product of an anteroposteriorly elongate pterygoid-pterygoid contact (as in Rhynchosauria).
As such, we have eliminated that character from this study to avoid redundancy. -The original formulation of this character presented in Pritchard et al. [68] and Nesbitt et al. [10] described the contact between the paroccipital process and the squamosal. However, these codings ignored taxa in which the lateral tip of the paroccipital process contacts the suspensorium at the supratemporal or quadrate [156], depending on the relative development of those elements. For the moment, we retain a single character to describe the presence or absence of a paroccipital process-suspensorium contact, hypothesizing that such a contact is homologous across Diapsida. -This character expands on the original Character 59 in Pritchard et al. [68] and Nesbitt et al. [10], which only described the presence of the dorsomedially inclined processes of the exoccipitals. In those studies, Character 60 described the presence or absence of a supraoccipital contribution to the dorsal margin of the foramen magnum. However, in studying those taxa that lack an exposure of the supraoccipital on the foramen magnum, we have recognized no taxa that lack dorsomedial processes that do not exhibit a supraoccipital exposure. As the incipient presence of these processes appears to be a necessary intermediate condition between the columnar exoccipitals and the complete exclusion of the supraoccipital from the foramen magnum, we elected to combine the original characters into a single, ordered character for this study. -This character was present in Pritchard et al. [68] and Nesbitt et al. [10] as Character 61.
-This character was modified from Character 60 in Pritchard et al. [68] and Nesbitt et al. [10], which described the presence of fusion between exoccipital and opisthotic. We have noted the presence of numerous taxa that exhibit a distinct fusion between the basioccipital and exoccipital (e.g. CQ drepanosaurid (AMNH FARB 30834), Pr. broomi (BP/1 2675), Czatkowiella harae [56]), for which we have introduced an additional state.
-This character was present in Pritchard et al. [68] and Nesbitt et al. [10] as Character 62.

63) Basioccipital, occipital condyle, posterior surface: (0) exhibits elliptical notochordal depression that occupies much of posterior surface of condyle; (1) exhibits narrow 'pinprick' notochordal pit within posterior surface; (2) condyle is smoothly convex. ORDERED.
-We introduce this character here to describe both the presence of a notochordal pit within the occipital condyle and the relative development of that pit. As early Sauria exhibit a range of morphologies, including exceptionally broad pits and extremely transversely narrow pits, we include an intermediate state between an 'unpitted' condyle and very broad, prominent pits.
-The original Character 64 in Pritchard et al. [68] and Nesbitt et al. [10] described the presence/absence of basioccipital basal tubera. In examining the early sauropsid and diapsid taxa purported to lack such tubera (e.g. Araeoscelis, Captorhinus), we recognized that the absence of such tubera in these taxa was more accurately described as a weak/incipient development of the tubera in which they did not extend far ventrally relative to the occipital condyle. The character has been rephrased to account for this morphological detail. -We have added further morphological description than present in Pritchard et al. [68] or Nesbitt et al. [10] to this character to clarify how we define 'crista prootica'.

74) Prootic, anteroventral surface, anterior inferior process: (0) present, framing anterior margin of trigeminal foramen; (1) absent, trigeminal foramen unframed anteriorly.
-We have added further morphological description than present in Pritchard et al. [68] or Nesbitt et al. [10] to this character to clarify how we define 'anterior inferior process'. -This character is slightly modified from Character 87 of Pritchard et al. [68] and Nesbitt et al. [10], which described the fusion of articular and prearticular as a character state describing the 'composition' of the retroarticular process. In considering the condition of most Lepidosauria, which bear this fusion, it is more appropriate to simply describe the fusion of these two elements as the character. -This character has been modified from Character 94 in Pritchard et al. [68] and Nesbitt et al. [10], incorporating two states for what was regarded as acrodonty (superficial attachment of teeth to dentigerous bones) in the original datasets. The new states describe intermediate stages of acrodonty, as revealed by CT scanning of early rhynchocephalians (described in [157]).
In Planocephalosaurus robinsonae and Diphydontosaurus avonis, teeth appear to be superficially attached to their respective dentigerous bones, but with pulp cavities that extend into said bones. By contrast, in Sp. punctatum and Cle. hudsoni, the pulp cavities do not invade the dentigerous bones. We order this character under the hypothesis that the reduction in the extent of the pulp cavity as an intermediate condition between rooted teeth and teeth with entirely superficial attachment. Taxa that lack CT investigation of pulp cavity morphology we code as 1 and 2. -We removed the original Character 99 from Pritchard et al. [68] and Nesbitt et al. [10], which described palatal dentition morphology (small, button-shaped teeth versus conical teeth). These poorly defined states require further study, especially of small-bodied taxa on which such characters are difficult to discern (e.g. Pe. kansensis, Y. capensis). Note that Character 48 describes morphological distinctions in the palatine dentition of Rhynchocephalia. -This character describes the substantial heterogeneity in tooth size in Rhynchocephalia. -We removed the original Character 101 from Pritchard et al. [68] and Nesbitt et al. [10], which described the relative concavity of the anterior articular surface of the vertebral centra. No taxa studied first-hand were coded as lacking an anterior concavity to the vertebral centrum; indeed, the only animals coded as such were based on text descriptions of vertebral morphology (e.g. [158] for Orovenator mayorum). In the light of the poor definition of this character and its absence in specimens studied first-hand, we elect to remove it from the study. -This character describes the midline canal in the vertebral centra of many early Diapsida and Lepidosauromorpha (e.g. G. bridensis [119], Cle. hudsoni [110] and Sp. punctatum (YPM R 10646)). -We have modified the definition of the second state of this character, which initially noted that epipophyses were required to be 'posteriorly projecting'. In some taxa (e.g. Az. madagaskarensis [10], Pa. dolichotrachela [130]), there are prominent ridges on the dorsum of the postzygapophyses, but these do not point posteriorly or posterodorsally. As such, we have modified the definition of epipophyses to be more inclusive. Note that Character 271 in this analysis describes the shape and development of these epipophyses. -The original Character 126 in Pritchard et al. [68] and Nesbitt et al. [10] described the degree of transverse expansion in the dorsal neural spines. We chose to eliminate that character, as it did not account for the nature of that expansion (e.g. whether or not is was formed by a transverse broadening of the bone of the tip of the spine, or if it was formed by mammillary processes just distal to the tip). Further study is definitely needed on vertebral variation in early Diapsida, but for the moment we describe the presence of these processes in the novel character above.

136) Chevrons, hemal spine, shape: (0) tapers along its proximodistal length; (1) broadens slightly along its length; (2) broadens distally, forming inverted T-shape broadens distally forming subcircular expansion.
-Pritchard et al. [68] and Nesbitt et al. [10] included a state that described chevrons that maintain their anteroposterior breadth along their proximodistal length. This is not strictly the case in most taxa that were coded as such in these analyses; in most early archosauromorphs (e.g. Protorosaurus speneri, Mes. browni), the chevrons broaden very slightly at their dista ends. The codings have been changed as such.

137) Chevrons, hemal spine, length: (0) similar in length or shorter than caudal neural spines, (1) substantially longer than caudal neural spines.
-In Pritchard et al. [68] and Nesbitt et al. [10], Character 137 described the presence and absence of gastralia. In Nesbitt et al. [10], Character 237 introduced a character that described the presence of abundant gastralia or their absence or extremely limited ossification. We elected to eliminate the original iteration of Character 137 for this analysis, as Character 237 is more descriptive. -We introduce this character to describe the extreme proximodistal elongation of the chevrons in certain early diapsid clades, especially drepanosauromorphs (e.g. Hy. limnaios (AMNH FARB 7759), D. unguicaudatus (MCSNB 5728)).

138) Chevrons, hemal spine, curvature: (0) roughly straight, (1) convex anteriorly.
-Pritchard et al. [68] and Nesbitt et al. [10] incorporated a character describing the number of pairs of lateral gastralia in each segment. Merck [92] described the presence of two pairs in most non-archosauriform diapsids, and a single pair in Archosauriformes. However, we note that two gastral ossifications are present on each side in coded early archosauromorphs (e.g. L. pandolfii, MFSN 1921; Ta. longobardicus, MCSN BES 1018; Proterosuchus alexanderi, NMQR 1484) and early rhynchocephalians [144] for which the character can be addressed. As no variation in this character can be addressed in this analysis, we exclude it from this iteration. -In Pritchard et al. [68] and Nesbitt et al. [10], this character includes the first two states listed here. We introduce a third state here to describe scapular blades that curve anterodorsally, as in drepanosauromorphs (e.g. Me. preonensis, MFSN 1769; V. cenensis, MCSNB 4751). -In the initial formulation of this character, Pritchard et al. [68] and Nesbitt et al. [10] only describe the presence and absence of the lateral pubic tubercle as described by [55]. Further study suggests that this rounded structure is likely homologous to the flattened ambiens muscle attachment in many early Sauria -In Pritchard et al. [68] and Nesbitt et al. [10], this character included fusion to the astragalus as a part of state 0. We have removed this statement as it adds an assumption that can only be supported through ontogenetic series or histological work on diapsid astragali. -Added terminology from [63]. -Character modified from Nesbitt et al. [10], which only described states for cervical centra with lengths greater than heights and heights greater than lengths. Ezcurra [63] described additional states for the relative lengths of anterior cervical centra. I have provided these states with consideration for the ratios apparent in the sample of taxa present in this specific analysis. -This character was included in Nesbitt et al. [10] along with Character 137 the original gastralia presence/absence character of Pritchard et al. [68]. The Pritchard et al. [68] character is removed in this study to avoid overprinting. 239) Astragalus, margin between tibial and fibular facets: (0) grades smoothly into anterior hollow of astragalus, (1) prominent ridge separates margin from anterior hollow.

240) Proximal tarsals, morphology of perforating foramen: (0) broad, marked by finished bone on astragalus and calcaneum; (1) pinched, marked by extremely constricted space between astragalus and calcaneum.
-The original Character 140 in Nesbitt et al. [10] described the anterior process of the chevrons in some early archosauromorphs (e.g. Tr. buettneri). However, this morphology was also incorporated into Character 130 of Pritchard et al. [68], resulting in overprinting of this morphology. We remove the character in Nesbitt et al. [ Pritchard et al. [68]) in contrast to the well-defined opening between the tarsals in most other taxa (e.g. Pr. broomi, BP/1 2676; Proterosuchus sp., AMNH FARB 2237). -Terminology added from Ezcurra [63] for the various processes of the dentary. -The remaining characters in this analysis are novel additions to the Pritchard et al. [68] and Nesbitt et al. [10] datasets. Some of these characters and codings are derived from Pritchard et al. [11]. -This character describes the relative anteroposterior length of the palatal process of the premaxilla, which is extensive in some early archosauromorphs. In South African Proterosuchus (e.g. NMQR 880, 1484) and 'Ch.' yuani (cast of IVPP V 36315), the palatal process of the premaxilla extends posteriorly past the premaxillary tooth row, paralleling the lateral margin of the maxilla. The palatal process in most archosauriforms is subequal in length to the dentigerous portion of the premaxilla (e.g. B. kupferzellensis [98], E. africanus (NHMUK R3592)).
-In most early archosauromorphs and non-saurian diapsids, the lateral foramen for the facial nerve is situated posterior to the crista prootica on the flat, lateral surface of the prootic (e.g. Az. madagaskarensis, FMNH PR 2765; Cz. harae [56]). In a number of archosauriforms, the crista prootica is appressed posteriorly to an additional crest of bone. This has the effect of 'sandwiching' the facial nerve foramen between these two bone crests. This condition may be seen in Proterosuchus fergusi (BP/1 3393), E. africanus (NHMUK R 3592), Fugusuchus hejiapenensis [63,168] and Xilousuchus sapingensis [168,169].
-In those early amniotes that bear teeth on the parasphenoid rostrum, the teeth run anteroposteriorly along the ventral surface of the structure, as well as being clustered at the base of the process. This condition occurs in Paleothyris acadiana [170], Pe. kansensis [131] and Lanthanolania ivakhnenkoi [171].
-The retroarticular process in most early Sauria exhibits a dorsally concave basin that likely contributed support for the tympanic membrane. In most early saurians, this basin is similar in anteroposterior elongation to the glenoid fossa itself (e.g. Tr. buettneri (TMM 31025- -In many early diapsids and lepidosauromorphs, the retroarticular process is dorsoventrally short, with its dorsal margin sloping ventrally relative to the quadrate articulation of the articular. This condition occurs in Y. capensis (AMNH FARB 5561), Me. preonensis (MFSN 1769), Cla. germaini [106] and Sp. punctatum [145]. By contrast, the retroarticular process of the CQ drepanosaurid and most archosauromorphs is dorsoventrally higher, with its dorsal margin at a level equivalent to the quadrate articulation. In archosauromorphs, this condition occurs in -Derived from similarly informative characters in Dilkes [33], Character 75; Ezcurra et al. [173], Character 167; and Ezcurra [63], Character 284.
-In Weigeltisauridae, the ribs of the dorsal region contact an additional, proximodistally elongate ossification that likely formed the framework for the patagium. This is seen in W. jaekeli [165] and Co. elivensis (MNHN MAP 327). These structures have not been reported from the isolated remains of Rautiania (e.g. [162,176]).
-In Kuehneosauridae, the ribs of the dorsal region are straight, splaying out from the midline. This construction is hypothesized to form the patagium (e.g. [123,124]) and may be seen in Kuehneosaurus spp. [124] and I. siefkeri (AMNH FARB 2101).
-Derived from similarly informative characters in Gauthier et al. [70] Character 99, which describes the unique articulation between radius and radiale in squamates. A very similar character state is evident in Chinle Formation Drepanosaurus sp. [11].
299) Second manual ungual: (0) similar in morphology to other manual, (1) substantially taller and more massive than other manual unguals.
-In a most early diapsids and Lepidosauria, the anterior margin of the iliac blade anterodorsal to the acetabulum is oriented posterodorsally. This condition occurs in Pe. kansensis [131], Th. colcanapi (MNHN MAP 360) and Sh. crocodilurus [138]. In some early archosauromorphs, the anterior margin of the iliac blade is vertically oriented (e.g. Tr. buettneri (TMM 31025-73)). In most early archosauromorphs, the anteroventral base of the iliac blade curves anterodorsally, providing a ventral base for an anterior expansion of the blade (e.g. Ta. longobardicus (PIMUZ T.1277), M. bassanii (MCSN V 457) and 'Ch.' yuani (IVPP V4067)). -Note that this character is independent of the presence of an anterior tuberosity on the anterior surface of the ilium. Such a tuber is present on the anterior margin of the bone in Sh. crocodilurus [138] and Tr. buettneri (TMM 31025-73), although this structure does not alter the orientation of the anterior base of the iliac blade.  -Derived from Ezcurra [63] Character 465, although we only incorporate a single state to describe this structure. We also refrain from describing this ridge as an attachment for M. caudifemoralis brevis pending further study of this ridge in early archosauromorphs that possess it (e.g. Ta Figure 10. Strict consensus of the analysis constraining Drepanosauromorpha and Tanystropheidae as sister taxa. (e.g. D. unguicaudatus (MCSNB 5278), CQ drepanosaurid (AMNH FARB 30834)), the neural arches of anterior cervical vertebrae are transversely broader than their respective centra.

C.4. Methods for phylogenetic analysis, topological constraint analyses and results
All analyses are run in TNT v. 1.5 [177], employing the 'Traditional Search' options including 10 000 replicates of Wagner trees (using random addition sequences), followed by tree bisection and reconnection (TBR) holding 10 trees per replicate. The best trees obtained at the end of the replicates were subjected to a final round of TBR branch swapping. We employed Rule 1 of Coddington & Scharff [178] for collapsing zero-length branches. We designated the Carboniferous diapsid Pe. kansensis as the outgroup for this analysis. We employed the STATS.RUN TNT script to obtain the Consistency Index and Retention Index for all trees. We used the Bremer support option in the Trees submenu of TNT, calculating supports based on a new round of TBR, holding trees suboptimal by 15 steps. A nexus file containing the phylogenetic data matrix analysed in this paper is available on Data Dryad [179]. The matrix is also published on Morphobank (www.morphobank.org) as project 2214.  Figure 11. Strict consensus of the analysis constraining Drepanosauromorpha and Protorosaurus speneri as sister taxa.
The primary analysis presented in figures 3 and 4 recovered four most-parsimonious trees of 1009 steps in length (CI = 0.336, RI = 0.651) recovered in 4277 out of 10 000 replicates. We also ran a jackknife analysis (10 000replicates, 20% character removal probability), the results of which are presented as GC values (frequency differences). The results of this analysis are presented in figure 8. A stratigraphically calibrated strict consensus is presented in figure 9.
To test the suboptimality of past hypotheses of the affinities of Drepanosauromorpha among Diapsida, we ran several constraint analyses. In each case, we created unresolved trees in which the only resolved clades included (1) all included drepanosauromorphs in an unresolved polytomy and (2) the hypothesized sister clade (or species) in an unresolved polytomy. Our constraint analyses included the following permutations: (i) a sister relationship with Tanystropheidae (hypothesized by [33]), (ii) a sister relationship with Protorosaurus speneri (hypothesized for Me. preonensis in [34]), (iii) a sister relationship with Weigeltisauridae (hypothesized in an analysis including Coelurosauravus and most drepanosauromorph taxa by [36]) and (iv) a sister relationship with Kuehneosauridae (hypothesized by [94]). For each constraint analysis, we used the same search methods described above for the unconstrained analysis.  than the most-parsimonious trees in the primary analysis. This analysis recovers a monophyletic Protorosauria in the sense of Dilkes [33]; Protorosaurus speneri is the sister taxon of the clade Tanystropheidae + Drepanosauromorpha. Archosauromorpha is substantially less resolved than in the primary analysis. The resulting topology is presented in figure 10. -Enforcing a sister relationship with Protorosaurus speneri produced six most-parsimonious trees of 1024 steps in length recovered in 8738 out of 10 000 replicates. This result is 15 steps longer than the most-parsimonious trees in the primary analysis. In the mostparsimonious trees, Protorosaurus + Drepanosauromorpha is the earliest diverging branch within Archosauromorpha following the Archosauria + Lepidosauria divergence. The phylogeny of Archosauromorpha has altered, with Boreopricea + Kuehneosauridae being the next divergence, followed by Allokotosauria. The phylogeny of non-saurian diapsids is largely congruent with the primary analysis, although Claudiosaurus is now the earliest diverging diapsid rather than Orovenator. The resulting topology is presented in figure 11. -Enforcing a sister relationship with Weigeltisauridae produced 32 most-parsimonious trees of 1011 steps in length recovered in 46 238 of 10 000 replicates. This result is two steps longer than the most-parsimonious trees in the primary analysis. In the resultant topologies, non-saurian diapsids are poorly resolved. The topology of Sauria is identical to the primary analysis. The resulting topology is presented in figure 12.  Figure 13. Strict consensus of the analysis constraining Drepanosauromorpha and Kuehneosauridae as sister taxa.
-Enforcing a sister relationship with kuehneosaurids produced two most-parsimonious trees of 1019 steps in length recovered in 9547 out of 10 000 replicates. This result is 10 steps longer than the most-parsimonious tree in the primary analysis. In the most-parsimonious trees, Drepanosauromorpha and Kuehneosauridae are the sister taxon of Protorosaurus + all other Archosauromorpha. The phylogeny remains otherwise similar in its topology, although Claudiosaurus is now the earliest-diverging diapsid after Pe. kansensis. Boreopricea funerea is the sister taxon of Prolacerta + Archosauriformes. The resulting topology is presented in figure 13.